Hi Andrew, thank you very much. I am really amazed how fast and friendly the responses are on this list.
On Wed, 2011-07-13 at 13:12 +0200, Andrew McBride wrote: > Hi Johannes > > > > The result however > > is not what I want. > > The results are correct though. Think of two one-dimensional linear elements > of unit length. Integrate a constant function f over them and them. Then \int > f dx = 2f but the vector of nodal f will be [f/2, f, f/2] > > > It is not constant and at the boundaries the value > > of the reconstructed function is too small. I believe this is the case, > > because there are contributions to the value of the boundary dofs are > > implicitly zero, as the cells for these contributions are not existing. > > no this is not correct. it's just that the supports of the functions are > truncated at the boundary. Think of the 1D example again. The shape functions > are nodally defined and are thus truncated on the boundary. I think what you say is exactly what I tried to say, just using the proper words. > > > What is an appropriate way to do that? > > If you really want to scale based on the patch area of the nodal shape > function then you could follow the approach in Hughes, The Finite Element > Method, pg 229 where the shape functions are weighted as you suggest. Thank you, I will try to get my hands on that. Regards, Johannes > > > Cheers > Andrew > > > > > Thank you in advance > > > > Regards > > > > Johannes Reinhardt > > > > > > > > > > > > > > <free_boundary.cpp><notconstant.png>_______________________________________________ > > dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii > _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
